Applications of spectral inversion to understanding vegetation functional trait relationships

Abstract:

Spectral data from both field observations and remote sensing platforms are a rich source of information for studying plant traits. Traditional approaches to using spectral data for studying vegetation have proven effective in sensor-, site-, or plant type-specific settings, but differences in model assumptions and failure to account for uncertainties have hindered efforts to synthesize observations from different sources and use spectral data in a predictive capacity. Here we present a novel approach that uses Bayesian inversion of the PROSPECT 5 leaf radiative transfer model (RTM) to investigate the ability of spectral data to inform our understanding of plant functional traits. First, we validated our method by comparing inversion results to independent measurements of relevant leaf structural and biochemical parameters. Second, we tested the accuracy and precision of RTM parameter retrieval as a function of spectral resolution and quality by performing inversions on simulated observations for a variety of common remote sensing platforms. We observed predictable increases in parameter uncertainty and covariance with declining spectral resolution, but we also found that the measurement characteristics of all sensors are capable of providing information about at least some of the parameters of interest. Finally, we applied our inversion to a large database of field spectra and plant traits spanning tropical, temperate, and boreal forests, agricultural plots, arid shrublands, and tundra to identify dominant sources of variability and characterize trade-offs in plant functional traits. We found substantial intraspecific variability in traits and explored the extent to which this variability falls along the same axes as the interspecific leaf economics spectrum. Ultimately, our results show that Bayesian RTM inversion provides a powerful framework for using spectral data to inform our understanding of plant functional traits and how they are linked with ecosystem processes.